TAUS - Translation Automation

According to the results of the March 2009 TAUS market survey, only 14% of the LSP (language service provider) respondents state that they will never use MT, 40% already use it today. This provides strong evidence that MT is moving into the mainstream among LSPS. To launch a series of reports on LSP deployment, this report focuses on the two major LSPs which own and develop their own MT systems.

The world's second and third largest language service provider companies Lionbridge Technologies (US) (2008 revenues $419.0 million 3,820 employees, 48 offices) and SDL International (UK) (2008 revenues $174.5 million, 1,700 employees, over 50 offices), are also two of the very few LSPs to have acquired, developed and use their own machine translation systems.

Other LSPs who have some experience in deploying MT systems include CLS Communications (DE, CH), which uses an in-house version of the BrainTribe MT system (now developed and marketed under the Lucy Software umbrella), and WorldLingo (US), which uses MT technology as both an advertising gimmick and a production tool. Compared with them, however, SDL and Lionbridge maintain substantial development teams for their wholly-owned systems, but until very recently, have tended to use the technology as a discreet asset rather than as a major plank in their strategic or marketing development.

However there are clear differences in the two companies' MT agendas. Unlike Lionbridge, SDL also sells its translation technology, alongside its services. Having first acquired Trados (translation memory and terminology tools) and then Idiom (GMS solution), with their existing broad customer bases in the translation industry, SDL announced in October 2008 that it was marketing a new range of translation technology products which includes its "automated" translation solution. This brings SDL out of the MT as a production option space and into the MT vendor market where systems are sold either under license or as off the shelf products. It is therefore competing either head-on or indirectly with players such as Language Weaver (US), Systran, and PROMT (RU), as well as with the dozen or so much smaller technology providers working across restricted language pairs, (e.g. AppTek (US) and Linguatec(DE)). Like some other vendors, it also offers a free machine translation service online (http://www.freetranslation.com/) for the automatic translation of websites and short text.

This report looks at the MT strategies of these two emblematic LSP competitors at a time when the translation industry is experiencing a bout of radical innovation in technology, business and publishing models, along with workforce organization.

There is probably only a slight chance today of another LSP investing deeply in MT technology as a development option, more because of the very high cost of skilled human resources and the development/training process than the actual technology itself in a market that is under increasing price pressure.

However, TAUS has collected evidence of the strong likelihood that LSPs will engage much more closely with MT, either by licensing some form of technology, working in partnership with an MT vendor (as in the PROMT's LSP Partnership Program) or working more closely as a supplier of post-editing services with large-account customers who may decide to use the technology in-house.

It is also more likely for language-centric people such as translators to opt for RBMT rather than SMT systems. Rule-based systems embody declarative "linguistic knowledge", whereas data-driven language-independent SMT systems take a more "engineering" approach to the process.

However, a number of LSPs are beginning to experiment with the Moses open-source MT engine, which means that their principle costs will be in people rather than software licences. It may well be possible for even smaller LSPs to train and tune very finely-targeted SMT engines, once they have access to large collections of language data (e.g. via TDA), and build up capabilities in SMT deployment. Examples of LSPs going down this route are Translated.net and Pangeanic.

Lionbridge and SDL are therefore likely to remain in a special category as LSPs due to their capacity to develop MT technology as well as service clients with production solutions. At the same time, they are pursuing quite different strategies.

SDL continues to act as technology supplier, both at the level of infrastructure with its newly Common Enterprise Application Framework, and in terms of selling tools (MT system, authoring system, GMS, translation memory), much to dismay of many LSPs who feel locked into these products while SDL is competing against them for translation business. It will be interesting to see how many LSPs interested in deploying MT as a partial solution to their service requirements benefit from SDL's MT offering, rather than using pure plays such as PROMT or Systran, for example. LSPs who act as subcontractors to SDL already have access to the service as a option, and a small subset of these also license MT software from MT system vendors, playing off different quality levels and language pairs as they search for the optimum cost-time-quality equation.